@alexcooldev: Currently, there are a lot of ...

Currently, there are a lot of self-improvement apps using AI influencers to promote their apps on TikTok, and they’re going extremely viral. You can check out these AI influencer accounts.

And yeah, all of these TikTok accounts are promoting one app called Her 75. It’s 100% organic with no paid ads, and the app is making over $40k/month. (Like I mentioned before, those analytics platforms usually underestimate real revenue by around 50–150%, meaning the actual revenue could be closer to $60k–$70k/month with extremely high profit margins.)

And now I’m going to guide you step by step on how to create AI influencers like this using Arcads. (Why Arcads? Because Arcads is partnering with me and sponsoring unlimited credits so I can create bulk AI influencer videos. But you can absolutely use any AI tools that fit your workflow.)

Now let's get starting.

Step 1: Create highly realistic AI influencers using GPT Images 2 or Nano Banano with JSON format prompts.

Why JSON format instead of normal prompts? Because JSON gives you a clear structure and makes the output much more consistent. Instead of writing a random paragraph prompt every time, you can define the influencer’s face, hairstyle, outfit, camera angle, lighting, background, expression, and visual style in separate fields. This makes it easier to create multiple variations while keeping the same identity across different videos. It also helps AI tools understand exactly what you want and reduces random results.

JSON Image (worked very well with GPT Image 2):

```json
{
  "scene": {
    "type": "fitness_transformation_video_frame",
    "setting": "bedroom",
    "lighting": "warm indoor lighting",
    "camera_angle": "front-facing",
    "aspect_ratio": "9:16"
  },
  "subject": {
    "gender": "female",
    "pose": "standing with arms extended sideways",
    "expression": "neutral",
    "clothing": {
      "top": "black sports bra",
      "bottom": "white shorts"
    },
    "appearance": {
      "wet_hair": true,
      "body_type": "average/curvy"
    }
  },
  "background": {
    "wall_color": "light pink",
    "furniture": [
      "white wardrobe",
      "curtains",
      "floor vase with dried plants"
    ],
    "decor": [
      "minimal face sketch wall art",
      "stacked towels"
    ]
  },
  "style": {
    "aesthetic": "minimal cozy lifestyle",
    "platform": "TikTok/Reels"
  }
}
```

Step 2: Build the full AI video workflow inside Arcads

After creating the AI influencer image, I don’t want to jump between 10 different tools.

I want one workflow where I can create, test, remix, and scale the whole ad format.

That’s why I use Arcads as the main creative workspace.

Inside Arcads, I can use tools like GPT Image, Nano Banana, Kling, Seedance, Swap Actor, Camera Angle, Captions, Translate Video, Transcribe, Extract Frame, Extend Videos, Talking Actors, UGC Studio, and Create Workflow.

So instead of thinking:

“Let me generate one random AI video.”

I think:

“How do I build a repeatable content engine?”

For this case study, the goal is to recreate a TikTok-style fitness transformation video.

The format is simple:

Before body → after body → same person feeling → short text overlay → app as the system behind the transformation.

Something like:

“same person btw”

This format works because the hook is visual.

People don’t need to read a long caption.

They instantly understand the story in 1 second.

Step 3: Start with a proven reference video

The next thing I need is a real reference video.

This is important because I’m not asking AI to randomly invent movement.

I want to use a real TikTok motion pattern and apply it to my AI character.

For this type of video, I look for a reference that has:

The best reference videos are usually very simple.

A girl standing in a bedroom, extending her arms, turning slightly, and showing a fitness progress angle.

That’s it.

The more native the reference video feels, the better the final AI video feels.

Step 4: Use Kling Motion Control inside Arcads

Now I upload the reference video into Kling Motion Control inside Arcads.

This is the key part.

The reference video controls the motion.

The AI character controls the identity.

So the workflow becomes:

Reference video + AI character image = realistic AI influencer video

This is much better than normal image-to-video because the movement already comes from a real video.

For this fitness transformation format, that matters a lot.

Small details like breathing, arm movement, body posture, and phone-camera motion are what make the video feel real.

Step 5: Match the AI character to the reference video

This is where most people fail.

They upload a reference video with one pose, then upload an AI image with a totally different pose.

Then the output looks weird.

For better results, I try to match:

For example, if the reference video shows a girl wearing workout clothes with her arms extended, my AI character should also be wearing workout clothes with a similar pose.

The closer the AI image is to the reference video, the cleaner the motion transfer becomes.

Step 6: Keep the prompt simple

Since the motion already comes from the reference video, I don’t need a complicated prompt.

I only use the prompt to guide realism and avoid common AI mistakes.

Example prompt:

“Create a realistic vertical TikTok fitness transformation video using the uploaded reference video as the motion guide. Replace the original person with the uploaded AI character. Keep the same body-check pose, camera framing, and casual phone-camera style. Natural blinking, subtle breathing, realistic skin texture, stable body proportions, no face distortion, no body distortion, no extra fingers, no unnatural hands.”

Simple prompts usually work better here.

The goal is not cinematic perfection.

The goal is TikTok-native realism.

Step 7: Generate multiple versions, not just one

Never generate only one video.

AI video is a numbers game.

Sometimes the first output has weird hands.

Sometimes the second output has an unstable face.

Sometimes the third output looks fake.

Then version four suddenly looks almost perfect.

That’s why I generate multiple variations inside Arcads.

Same reference video.

Same AI character.

Same basic prompt.

Different outputs.

Then I pick the most realistic one.

This is the real workflow:

Generate in bulk → review → select the best → remix → scale.

Step 8: Create both “before” and “after” versions

For this transformation format, I need two AI character assets.

The first one is the “before” version.

This should look like a normal beginner fitness body.

Relatable, realistic, and not too perfect.

The second one is the “after” version.

Same person, but leaner, more toned, and more confident.

The important thing is identity consistency.

The face, hair, outfit, pose, camera angle, and lighting should feel similar.

If the after version looks like a completely different person, the video loses trust.

The transformation should feel believable.

Not extreme.

Believable.

Step 9: Run both versions through the same motion workflow

Now I run both images through Kling Motion Control.

First:

Reference motion + before character = before clip

Then:

Reference motion + after character = after clip

This gives me two clips with similar movement.

That’s what makes the transformation feel stronger.

The viewer sees the same pose, same motion, same person feeling, but a different body result.

That creates the “same person btw” effect.

Step 10: Use Arcads to create more variations

Once I have the core video, I can use other Arcads tools to create more variations.

For example:

I can use Captions to add subtitles.

I can use Translate Video to localize the creative into other markets.

I can use Swap Actor or Replace Actor to test another AI influencer.

I can use Camera Angle to create a different visual version.

I can use Extract Frame to pull strong still images from the video.

I can use Extend Videos if I want a longer version.

I can use UGC Studio or Talking Actors if I want to turn the same concept into a talking-style creator ad.

This is why Arcads is useful.

It’s not just one AI video tool.

It becomes the creative production system.

One idea can quickly become many ad variations.

Step 11: Create multiple hooks from the same visual

The visual format can stay the same, but the hook changes everything.

For example, I can test:

“same person btw”

“75 days later…”

“this is what consistency did”

“she didn’t need motivation, she needed a system”

“I stopped restarting every Monday”

“POV: you finally stayed consistent”

“not a diet, just a system”

Same video style.

Different hook.

Different audience psychology.

This is how I find the winning angle.

Most people try to create a completely new video every time.

But the better approach is to find one working format, then remix the hook, character, CTA, and app angle.

Step 12: Add the app as the reason behind the result

The app should not feel like a random ad at the end.

It should feel like the system behind the transformation.

For example:

“She didn’t need more motivation. She needed a system.”

Then show the app:

This makes the app feel natural.

The structure is:

Visual transformation → curiosity → app as the system → CTA

That’s much stronger than just saying:

“Download this app now.”

Step 13: Export the best versions and prepare them for posting

After generating and testing enough variations inside Arcads, I export only the strongest versions.

Not every output is worth posting.

I only keep the ones that look:

At this point, the core production is done.

Arcads helped me create the AI character workflow, apply real motion with Kling Motion Control, generate multiple variations, test different hooks, and prepare the final creative assets.

Then I can move into the final editing / posting stage.

Final workflow:

This is the real power of AI influencer content.

It’s not about creating one AI video.

It’s about building a repeatable creative engine.

One viral format can become 10, 20, or 50 different variations.

And when you combine AI characters, real reference motion, Kling Motion Control, and Arcads workflow, you can create TikTok-native AI influencer videos at scale without hiring UGC creators for every single test.

This post is sponsored by Arcads AI, so you can still look for similar tools that fit your needs. 💪

Be like this guy, he made a one-feature funny Mac app just for entertainment, got 2 viral videos, and made $44k in 60 days. I’ve honestly never seen a case like this before 🥴

The new formula for a successful product in 2026:
Build an app -> go viral on TikTok & IG -> money printer 💸
x.com/robj3d3/status…